Manuscript
A repetition-suppression account of between-trial effects in the Stroop task
Ion Juvina
Department of Psychology
Baker Hall, 336A
Carnegie Mellon University
5000 Forbes Avenue, Pittsburgh PA 15213
Tel. +1-412-268-2837
Email: [email protected]
Niels A. Taatgen
Department of Psychology
Baker Hall, 345B
Carnegie Mellon University
5000 Forbes Avenue, Pittsburgh PA 15213
Tel. +1-412-268-2815
Email: [email protected]
Short title: A repetition-suppression account
Send correspondence to: Ion Juvina, [email protected]
Manuscript
Abstract
Some of the influential theories and models of cognitive control are challenged when
attempting to explain empirical data that replicate previously reported findings. These
findings refer to four between-trial effects in the Stroop task, previously described in the
literature on Stroop effects, negative priming, and inhibition of return. Existing
theoretical accounts are shown to fail to explain all these four effects in an integrated
way. A repetition-suppression mechanism is put forward in order to account for these
data in an integrated way. Two computational cognitive models implementing this
repetition-suppression mechanism are proposed. The values and limitations of each of
these models, as well as their theoretical implications, are discussed.
Manuscript
Acknowledgments
This work was funded by the ONR grant no. N00014-06-1-0055. The authors thank
Daniel Dickison for adjusting the finst mechanism in ACT-R.
Manuscript
1. Introduction
Recent advances in Cognitive Neuroscience have made possible to better understand the
nature and mechanisms of cognitive control. One of the typical functions of cognitive
control is to compose task-relevant temporal sequences of actions. This paper attempts to
prove with empirical and computational arguments that repetition-suppression is
sometimes employed in temporal sequencing of actions, particularly when repetition of
stimuli complicates the process of response selection.
It is currently under debate whether suppression (also referred to as cognitive inhibition)
is one of the mechanisms of cognitive control. Some authors assert that cognitive
inhibition is essential for cognitive control (Aron, 2007; Houghton & Tipper, 1996),
others say that it is unnecessary. In most (if not all) of the cases, the behavioral effects
attributable to cognitive control can be accounted for equally well with or without a
suppression mechanism (Egner & Hirsch, 2005; MacLeod, Dodd, Sheard, Wilson, &
Bibi, 2003).
In order to disentangle concurrent accounts we impose several methodological
constraints: a viable theoretical account should be able to simultaneously explain a large
range of effects, it should be possible to be expressed in computational terms, and it
should be biologically plausible.
Manuscript
The classical Stroop task will be used because it is a landmark task for studying cognitive
control (Miyake et al., 2000) and an extensive body of literature has accumulated over
many years to endorse the robustness of the behavioral effects as well as to aid with
understanding the cognitive mechanisms responsible for these effects (MacLeod, 1991).
However, the theoretical arguments put forward in this paper are not limited to the Stroop
task; they have some bearing upon a whole range of tasks, particularly tasks involving
processing temporal sequences of stimuli and selecting amongst potentially conflicting
responses (what is sometimes referred to as “the continuous performance paradigm”).
The following section presents replications of known effects in the Stroop task and
estimations of their relative magnitudes and time course. Section 3 shows how this
empirical data challenge established theories and models of cognitive control and
presents an alternative account. Section 4 presents two computational models that
implement this alternative account. Section 5 discusses the contributions of each of these
models to a coherent theory of cognitive control.
2. Empirical studies
The empirical research presented here is part of an ongoing project aimed at contributing
toward a coherent theory of cognitive control. In one of the most comprehensive reviews
of research on the Stroop task, MacLeod (1991) described a series of between-trial effects
and proposed a suppression mechanism to account for all of them:
Manuscript
“When the irrelevant word on trial n-1 is the name of the target ink color on trial
n, interference with color naming will be enhanced temporarily; when the ink
color on trial n-1 matches the word on trial n, there will be some facilitation of
color naming on trial n. If the word on trial n-1 is repeated on trial n, then the
word is already suppressed and will cause less interference in naming a different
ink color on trial n. An interesting study would be to mix these two types of
repetition effects in the same experiment, directly comparing their size.”
“My own bias […] is to invoke a suppression idea so that the facilitation and
interference effects as a result of item sequence have a common grounding”
(MacLeod, 1991, pp. 178)
It was our intention to conduct such an “interesting study” in which to replicate all these
between-trial effects and investigate whether a single integrated account can explain all
of them as suggested by MacLeod in 1991. However, MacLeod has recently advocated
against cognitive inhibition as an explanatory mechanism for attention and memory
phenomena including negative priming and inhibition-of-return (MacLeod, 2007a;
MacLeod, 2007b; MacLeod, Dodd, Sheard, Wilson, & Bibi, 2003). Thus, the research
question we are trying to address here is whether this integrated account should be based
on suppression (cognitive inhibition) or not. A subsequent goal will be to use neurally
plausible computational modeling in order to understand how a cognitive control
mechanism might operate in the human brain.
There were two empirical studies. The first study replicated all three between-trial effects
described in the quotation above (for brevity, they will be referred to as Word-Color,
Manuscript
Color-Word, and Word-Word, respectively), but also revealed a significant Color-Color
effect, which was also a replication of a known effect, though previously reported in a
different context (Law, Pratt, & Abrams, 1995). This additional result motivated a second
study to replicate and increase the statistical power of the first study. Since the two
studies yielded very similar results (see section 2.5.3.), they will be presented together.
2.1. Participants
There were 53 participants in the first study self selected from Carnegie Mellon
University’s community. Age ranged from 18 to 59 with an average of 24. There were 16
women and 37 men. In the second study, there were 39 participants self selected from the
same pool, with age ranging from 18 to 54 and averaging at 25. There were 22 women
and 18 men. All participants received a fixed amount of monetary compensation for their
participation.
2.2. Design
The standard Stroop task was adapted for screen-based administration. Stimuli were
presented on a computer screen one at a time. There were three within-subject conditions:
incongruent, congruent, and neutral. The three trial types corresponding to the three
conditions were randomly mixed (non-blocked). Trial order was randomized for each
participant. In the first study, every subject received 150 trials, 50 trials for each
condition. In the second study, there were 240 trials per participant, 80 trials for each
Manuscript
condition. The set size of neutral stimuli was decreased from 53 to 10. These changes
were made in order to ensure that each participant encountered at least 30 repetitions for
each of the between-trial effects. The location of stimuli on the screen was kept constant.
2.3. Materials
A computer-based variant of the Stroop task was designed for the purpose of this
research. Stimuli were color words (red, blue, yellow and green) and neutral words
colored with one of the four colors denoted by the mentioned color words. They were
presented one at a time and remained on the screen until the participant responded. Two
response options were also displayed flanking the stimulus on its left and right sides.
Response options were non-colored (i.e., colored with black ink) color words. One
response option would contain the right answer and the other one a wrong answer.
2.4. Procedure
Instead of verbally naming the color of the stimulus as in the classical Stroop task,
participants were instructed to select as fast as possible the response option that matched
the color of the stimulus from the two options presented on the left and right sides of the
stimulus by pressing a key for each option. The session started with a short computerguided tutorial that emphasized the correct response. During the actual task no feedback
was provided.
Manuscript
2.5. Results
Data of two participants were excluded from analysis. One of these participants had very
high reaction times, above 2000ms on average, and this criterion had been used before to
exclude data from analysis (Miyake et al., 2000). The other participant that was excluded
seemed to have misunderstood the task instructions. As he or she had an accuracy of zero
(minimum) for the incongruent condition and one (maximum) for the congruent
condition, we inferred that he or she reacted as if responding to the word dimension
instead of the color dimension of the stimulus. Aside from this, no other manipulation of
the data was done.
Reaction time (RT) was used as a dependent measure. Usually when RT is used as a
dependent measure it is log-transformed in order to correct for its skewed distribution.
We compared the results with and without the log-transformation of RT, and results were
similar. Thus, for presentation purposes, we decided to use the original (not transformed)
variable; this way the magnitudes of all effects are always expressed in milliseconds. It is
also a common practice to use only data from correct trials. However, as far as RT was
concerned, it did not make much of a difference if the analysis was performed on the
whole dataset (including the incorrect trials) as compared with the restricted dataset (only
the correct trials). We chose to present here the analyses performed on the whole dataset.
2.5.1. Within-trial effects
Manuscript
Accuracy data are consistent with previous studies, showing less than 2% errors for the
congruent and neutral conditions and less than 10% errors for the incongruent condition
(Table 1). Significant interference and facilitation effects were found. Since within-trial
effects were very consistent with those found in previous studies they will not be treated
in more detail here.
(Table 1 about here)
2.5.2. Between-trial effects
These are effects related to a particular sequence of trials. They will be described one by
one starting with the ones mentioned by MacLeod (1991). Since we intend to compare
various accounts, we will try to resist at this point to label these effects with terms that
might suggest particular explanatory mechanisms (negative priming, inhibition-of-return
etc.), as recommended by MacLeod (2003). Table 2 presents a synoptic with examples of
all these effects.
(Table 2 about here)
Repetition priming data, that is, cases in which the exact stimulus was repeated (for
example, the word RED in red ink repeated as such) were excluded from the analyses of
between-trial effects, for the following reasons: (1) reactions to repeated stimuli were
much faster than all the other reactions and, arguably, a different type of response
selection was involved, that is, a heuristic of the kind “if no stimulus change, then repeat
Manuscript
the previous response” (Klein, 2004); (2) it would be impossible to establish where such
a case belongs (for example, the congruent RED repetition mentioned above could be
classified as any of the four between-trial effects); (3) it would artificially inflate or
deflate the magnitudes of the between-trial effects depending whether these cases were
included in the prime trials or in the baseline (Christie & Klein, 2001). The number of
such cases was very small, that is, 4% of the total number of trials. Frequencies for the
between-trial effects were as follows: Word-Color 15%, Color-Word 14%, Word-Word
9%, and Color-Color 21% of the total number of trials. The absolute magnitudes of these
effects will be presented as averaged differences between reaction times on prime versus
non-prime trials (Table 3).
(Table 3 about here)
2.5.2.1. The Word-Color effect (negative priming)
When the word on trial n-1 names the color on trial n, reaction time increases. As shown
in Table 3, the highest increase occurs in the incongruent condition (102ms), followed by
the congruent condition (93ms) and neutral condition (40ms). This effect has been
replicated many times, is very robust and fairly general (see Tipper, 2001, for a review).
2.5.2.2. The Color-Word effect
Manuscript
When the color on trial n-1 is the same as the denotation of the word on trial n, reaction
time decreases. Notice that this effect cannot occur in the neutral condition, because the
neutral word does not denote any color. As shown in Table 3, the decrease in reaction
time occurs solely in the incongruent condition (-74ms); in the congruent condition there
a very small (practically zero) increase in reaction time (8ms). This effect was first
reported by Effler (1977) and replicated several times (see MacLeod, 1991, for a review).
2.5.2.3. The Word-Word effect
When the word on trial n-1 is the same as the word on trial n, regardless of the colors of
these words, reaction time decreases for the incongruent and neutral trials (-68ms and 51ms, respectively) and increases for the congruent trials (93ms) (Table 3). Notice that
this effect is reversed in the congruent condition, that is, word repetition reduces both
Stroop interference and facilitation. This result is a replication of (Thomas, 1977)
reviewed in (MacLeod, 1991)
2.5.2.4. The Color-Color effect
When the color on trial n-1 is the same as the color on trial n, regardless of the words of
these stimuli, reaction time increases. As shown in Table 3, the highest increase occurs in
the incongruent condition (79ms), followed by the neutral condition (23ms); the increase
is almost zero in the congruent condition (8ms). This effect has not been reported in the
Manuscript
context of the Stroop task (to our knowledge) but it was reported in the literature on
inhibition-of-return (Law, Pratt, & Abrams, 1995).
2.5.3. Time course and relative magnitudes of between-trial effects
In order to check for statistical significance these data were submitted to a Linear Mixed
Effects (LME) model. This type of analysis was chosen because it allows controlling for
individual differences and accurately determining the magnitudes of small effects in
hierarchically nested data. Thus, the data points corresponding to the within- and
between-trial effects are not independent across subjects; they are nested within subjects.
It is known that priming effects are rather small in magnitude, often around 20 ms
(MacLeod & Bors, 2002). Ignoring the inherent nesting characteristic of the data would
diminish or vanish some of the small-size effects.
A first analysis was intended to check if there are significant differences between the two
studies. A LME model with reaction time (RT) as dependent variable and study,
condition and the four between-trial effects as independent variables was run. Results
showed a main effect of study (t=2.038, p=0.045) and an interaction between study and
condition (t=2.41, p=0.016). Reaction time was smaller overall in the second study as
compared with the first study, and particularly in the congruent condition. Differences in
average reaction time between different studies have also been reported by MacLeod and
Bors (2002). There were no significant interactions between study and any of the four
between-trial effects. The Pearson correlation between the two studies, calculated on both
Manuscript
within- and between-trial effects, was r=0.99. For these reasons the data of the two
studies were combined and analyzed together. A similar procedure of combining two
studies, the second being a replication of the first, was employed in (MacLeod & Bors,
2002).
Additional variables were included in the model to study the time course of between trial
effects. Thus, for each effect, not only the 1-back version was included but also the 2and 3-back versions. For example, the Word-Color-2back represents the case where the
word on trial n-2 names the color on trial n. Dependent variable is Reaction Time.
Independent variables are Condition (Incongruent, Congruent, and Neutral), and the four
between-trial effects together with their 2-back and 3-back correspondents. The
interaction between Condition and the Word-Word effect was also included in the model
as an independent variable based on the observation that the Word-Word effect is
reversed in the Congruent condition (see section 2.5.2.3. and Table 3).
The results of the initial LME model are presented in Table 4. Notice that all the four
effects are significant (alpha = 0.05). Some of the 2-back effects and the interaction
between Condition and the Word-Word effect are also significant.
(Table 4 about here)
The coefficients of this LME model can be used as estimates of the magnitudes of the
between-trial effects. For example, for the Word-Color effect, the value of the LME
Manuscript
coefficient indicates an increase in reaction time for the prime trials as compared to the
non-prime trials of 50.7 ms, should all the other variables in the model be maintained
constant. The coefficients for all the between-trial effects are plotted in Figure 1.
(Figure 1 about here)
A stepwise procedure was employed in order to find the best LME model that fits the
data amongst alternative models. This procedure starts with the largest model (Table 4),
computes the Akaike Information Criterion (AIC) for each factor, and drops one factor at
the time until no factor can be dropped without a significant decrease in AIC. The best
model found for our data is summarized in Table 5.
(Table 5 about here)
The best fitting model has dropped all the 3-back versions of the between-trial effects but
has retained the 2-back version of the Color-Color effect even though it was only
marginally significant. It is noteworthy to observe that the estimates of the between-trial
effects are very robust, that is, they do not change (not much) their value from the initial
to the last model.
Since most of the 2-back versions of the between-trial effects were maintained in the best
model, the time course of the between-trial effects can be estimated as twice the average
duration of a trial (1069 ms), that is 2138 ms. Other studies have reported a longer time
Manuscript
course of between-trial effects (Erickson & Reder, 1998; Tipper, Weaver, Cameron,
Brehaut, & Bastedo, 1991).
A comprehensive exploratory analysis was also performed, starting with a LME model
that included all possible variables and their interactions and progressing toward the best
model in a stepwise manner as described above. The best model obtained this way was
not significantly different than the one presented above.
With regard to the magnitudes of the between-trial effects, the estimates used for model
simulations presented in section 4 will be the ones presented in Figure 1 and taken from
the initial LME model (Table 4). This was done in order to constrain the models to
simulate the original data in its entirety, that is, not only the statistically significant results
but also the results that indicate statistically nil effects. In addition, the estimates
presented in Figure 1 show a pattern of the data that might be indicative for the
underlying mechanism of the between-trial effects. Thus, Figure 1 suggests that betweentrial effects are maximal at 1-back and progressively decrease in magnitude at 2- and 3back. At 3-back their magnitude is statistically zero, but it seems that this is a result of a
gradual decay.
3. Discussion of the empirical results
Besides the well-known within-trial Stroop effects (increased and decreased RT in the
incongruent and congruent conditions, respectively, as compared to the neutral
Manuscript
condition), four significant between-trial effects have been found. Although not so well
known as the within-trial effects, these between-trial effects have been documented as
well (Law, Pratt, & Abrams, 1995; MacLeod, 1991). We have only replicated them and
estimated their relative magnitudes and durations while controlling for the within-trial
effects and individual differences. Although these findings have been known for a while,
to our knowledge, there is no integrated account for all four between-trial effects. This is
the first attempt that we know of to study all four effects together in a study and explain
them with a single account. An integrated suppression-based account has been suggested
in a review by MacLeod (1991), but recently the same author has expressed strong
criticism for any suppression-based account of attention and memory phenomena
(MacLeod, 2007a; MacLeod, 2007b; MacLeod, Dodd, Sheard, Wilson, & Bibi, 2003).
A first attempt to interpret these between-trial effects leads us toward a repetitionsuppression account: representations pertaining to just-completed trials are suppressed in
order to prevent them from interfering with future trials. However, we will resist for now
to advance such an account. As recommended by MacLeod (2003), we will first analyze
the available inhibition-free accounts.
The episodic retrieval account (Neill, 1997) holds that the to-be-named feature of the
current stimulus triggers an automatic retrieval (Logan, 1990) of the most recent episode
in which the concept corresponding to that feature has been used and the associated
reaction. For example, assuming the word “red” re-occurred as the color red, it would
trigger the retrieval of an episode composed of the concept “red” and the reaction “do-
Manuscript
not-respond”. Since the reaction derived from the retrieved episode is not adequate for
the current stimulus, an additional retrieval is required to generate the proper reaction,
which explains the time delay. This account predicts longer reaction times when the
previous word feature re-occurs as the current color feature, that is, the Word-Color
effect found in our data, also known as the negative priming effect. In the case of ColorWord repetition, this account would not predict a decrease in reaction time, as observed
in our data; the color feature of the preceding trial has the reaction “respond” associated
with it; when it comes back as the word feature of the current stimulus it should increase
interference because the irrelevant feature has now a “respond” reaction associated with
it. In the case of the previous color feature re-occurring as the current color feature (the
Color-Color case), this account would predict no increase in reaction times; the most
recent episode involving the current color contains exactly the reaction needed for the
current stimulus; thus, there would be no reason for an increase in reaction time as
observed in our data. In the case of Word-Word repetition, this account would probably
predict a decrease in reaction time (as observed in our data) because the previous word
feature has a “do-not-respond” associated with it, which could make it easier to reject the
same irrelevant word in the current trial. Thus, the episodic retrieval account predicts
well two of the four between-trial effects (Word-Color and Word-Word) but makes
wrong predictions for the other two effects (Color-Word and Color-Color).
Another inhibition-free account suggested by (MacLeod, Dodd, Sheard, Wilson, & Bibi,
2003) is the feature mismatch account (Park & Kanwisher, 1994). This account posits
that when the repetition is accompanied by a feature mismatch additional time is taken to
Manuscript
resolve this conflict. For example, when the “red” word precedes the red color, redness is
repeated but it occurs in conflicting features (word vs. color). This account predicts an
increase in reaction time for the Word-Color effect but also for the Color-Word effect
because there is a feature mismatch in this case as well. In the case of Color-Color
repetitions, this account would not predict an increase in reaction time because the feature
of the repeated entity does not change. There is also no feature mismatch in Word-Word
repetitions, thus an increase in reaction time would not be predicted. However, the feature
mismatch account would be unable to explain why there is a decrease in reaction time for
the Word-Word repetitions. In summary, the feature mismatch account explains only the
Word-Color effect, makes wrong predictions for the Color-Word and Color-Color effects,
and does not explain the Word-Word effect.
Since the Color-Color effect has been reported in the inhibition-of-return literature (Law,
Pratt, & Abrams, 1995) and interactions between Stroop effects and inhibition-of-return
effects have been documented (Fuentes, Boucart, Vivas, Alvarez, & Zimmerman, 2000;
Vivas & Fuentes, 2001), we have also considered the inhibition-free account of
inhibition-of-return suggested by (MacLeod, Dodd, Sheard, Wilson, & Bibi, 2003) and
called the attentional momentum account (Pratt, Spalek, & Bradshaw, 1999). According
to this account, attention can be oriented toward locations along the direction of
orientation faster than to locations that require a change in the direction of orientation.
This theory could not be applied to our case as such because all the stimuli appear at the
same location. However, object-based and semantic inhibition-of-return effects have been
documented (Fuentes, Vivas, & Humphreys, 1999; Tipper, Weaver, Jerreat, & Burak,
Manuscript
1994) and the attentional momentum theory can be extended to comprise, beside
locations, objects, and even abstract mental concepts such as redness. Provided we had
such an extended theory of attentional momentum able to account for all inhibition-ofreturn effects, would it explain our between-trial effects? Interestingly, this theory would
get very close to accounting for them all, if there were cues to redirect attention between
the Stroop trials as in the classical inhibition-of-return task (i.e., the cue-target paradigm).
However, the Stroop task was administrated in a continuous target-target paradigm.
There were no intervening stimuli (or cues) to redirect attention between two consecutive
trials. For example, in the Color-Color case, attention needs to be directed toward the
same concept as in the preceding trial, and its momentum should facilitate its orientation.
Similarly, this account fails to explain the other three effects as well: since there is no
change in the direction of attention with repetition, there would be no reason for
difficulty in re-orienting attention toward a just-attended entity. This account could only
explain the 2- and 3-back versions of the between-trial effects, because in these cases
there are intervening trials between repetitions.
If the existing inhibition-free theories cannot account for the presented data in an
integrated way, how about the existing accounts based on cognitive inhibition? One of
the most influential suppression-based accounts is the selective inhibition account
(Houghton, Tipper, Weaver, & Shore, 1996; Neill & Westberry, 1987). This account
posits an initial activation of both features (word and color) followed by an active
inhibition of the to-be-ignored feature (word) of the current stimulus. When the inhibited
feature returns as the to-be-named feature of the next stimulus, its inhibition has to be
Manuscript
overridden by re-activation. This account predicts longer reaction times when the
previous word feature re-occurs as the current color feature, that is, the Word-Color
effect, and shorter reaction time when the word repeats, that is, the Word-Word effect.
However, since only the to-be-ignored feature is inhibited (i.e., inhibition is selective),
this account predicts that reaction time will not increase when the previous color reoccurs as the current color. In fact, in the Color-Color case, reaction times should
decrease, since the to-be-named feature (color) has just been activated in the previous
trial. For the same reason, this account cannot explain the Color-Word effect. The color
of the preceding stimulus has been activated, thus when it re-occurs as the word of the
current stimulus, it has a higher potential to interfere with the color naming of the current
stimulus, causing reaction time to increase. In summary, the selective inhibition account
predicts well only two of the between-trial effects and it makes wrong predictions for the
remaining two. Thus, this account does not do any better that the inhibition-free accounts.
It seems that this account fails when it tries to explain between-trial effects as byproducts of within-trial effects, that is, when it posits that inhibition acts selectively at a
trial level in order to prevent the distractor from interfering with the target.
The account we propose, called repetition suppression, posits a control mechanism
dedicated to between-trial interference. It is the temporal sequencing of trials that this
control mechanism is used for and not the selection of targets from distractors. Temporal
sequencing of actions as a function of cognitive control is as important as the function of
selecting the relevant information out of the irrelevant one (Houghton & Tipper, 1996).
Manuscript
For the rest of the article we will use the term suppression instead of cognitive inhibition
in order to avoid the confusion between cognitive and neural inhibition, as recommended
by (MacLeod, 2007a). First, we will describe how repetition suppression can explain all
the between-trial effects in an integrated way. In the next section, we will present two
computational models that implement this account and we will discuss how the brain
could perform such a control function.
The repetition suppression account posits that at the end of a trial all representations that
have been used to make a decision in that trial are suppressed in order to prevent their
interference with the next trial. This suppression decreases in strength as the time passes
and it can be detected in behavior only when repetitions occur. In fact, this account
makes a stronger prediction: traces of this between-trial suppression should occur at any
time when repetitions occur. Thus, in the Word-Color effect, the concept denoted by the
word feature of the stimulus on the preceding trial re-occurs as the color feature of the
current stimulus. Since the representation of this concept has been suppressed, it takes
longer to name the color than in trials without this kind of repetition. In the Color-Word
effect, the word on the current trial has less potential to interfere with or to facilitate color
naming because its corresponding concept has been suppressed. This fact causes reduced
interference in the incongruent condition (i.e., decrease in reaction time) but also reduced
facilitation in the congruent condition (i.e., increase in reaction time) (see Table 3). In the
Color-Color effect, reaction time increases because the concept has just been suppressed
and it needs reactivation to be used in the current trial. In the Word-Word effect, the word
on the current trial has less potential to interfere with or to facilitate color naming in the
Manuscript
incongruent and congruent conditions, respectively. This explains why reaction time
decreases in the incongruent condition and increases in the congruent condition (see
Table 3 and the interaction between condition and Word-Word in the LME model).
The repetition suppression account resembles to a large extent the inhibition account of
inhibition-of-return (Posner & Cohen, 1984; Tipper, Weaver, Jerreat, & Burak, 1994).
What is different is that it operates at a semantic level. What is suppressed is not the
“return” of a particular representation of an object or location but the represented
concepts related to perceived features of stimuli, regardless whether these features are
targets or distractors (i.e., to be selected or to be ignored features). Repetition suppression
is a memory-based account explaining effects that occur in a continuous target-target
paradigm, and not an attention-based account explaining effects occurring in a cue-target
paradigm.
In summary, the repetition suppression account seems to be able to explain the betweentrial effects better than concurrent accounts and in an integrated way. The next selfimposed methodological constraint would be to implement this theoretical account in a
neurally plausible computational model.
4. Mechanisms of repetition suppression
As it is often the case, there is not a unique way to implement a theory at a computational
level. One possible way to constrain a computational model of a particular task is to look
Manuscript
at how the brain performs the same task. A review of the literature on inhibitory control
revealed two neurally plausible ways to implement a repetition suppression account in a
computational model. They will be respectively called bottom-up and top-down
suppression. Two computational cognitive models will be presented implementing these
two accounts, respectively. These models were developed with the aid of the last version
of the ACT-R cognitive architecture (ACT-R stands for Adapted Control of Thought –
Rational; Anderson, 2007). These two models are identical with regard to how they
implement the within-trial effects (see section 4.1) and they only differ with regard to
their implementation of the between-trial effects (sections 4.2 and 4.3).
4.1. The “relative automaticity” account
Many models of the within-trial effects have been developed (Cohen, Dunbar, &
McClelland, 1990; Herd, Banich, & O'Reilly, 2006; Lovett, 2005) and there seems to be a
large consensus that a “relative automaticity” account best explains these effects
(MacLeod & MacDonald, 2000). This account originates with an old finding in
psychology, namely that reading words is more automatic than naming colors (Cattell,
1886; Fraisse, 1969). This is explained by the fact that human adults have vastly greater
practice at reading words than at naming colors.
Our model of within-trial effects implements this difference in automaticity as difference
in strengths of association between representations of stimulus features and memory
elements associated with these features. Thus, when a stimulus is perceived, its word and
Manuscript
color dimensions are represented in the goal buffer – a storage structure used to maintain
information that is important for controlling the flow of actions involved by the task at
hand. For brevity, these representations will be referred to as control units. For example,
if the current stimulus is the word “blue” in red ink (incongruent condition), the two
control units are representations of the word “blue” and the color red, respectively. The
two control units spread activation toward associated memory elements biasing their
retrieval. The word “blue” spreads activation toward the concept of blueness, while the
color red spreads activation toward the concept of redness. The amount of activation
spreading from the goal buffer (called source activation, for brevity) is limited and is
equally shared by the two control units. The amount of activation received by a memory
element is a function of the amount of activation spread toward it and its strength of
association with the corresponding control unit. In our model, words have larger strengths
of association than colors. As a result, when a stimulus is presented, the concept
associated with its word dimension is more active than the concept associated with its
color dimension. In our example, blueness will be more active than redness. In order to
name the color of the current stimulus, a memory retrieval request is being made and the
concept of blueness is retrieved. At this point, if memory retrieval were sufficient for
performing an action, the model would commit an error, responding blue instead of red.
However, the behavior of an ACT-R model is guided not only by perception and memory
retrievals but also by firing of production rules of the kind “if condition, then action”. In
this case, a production rule detects the wrong retrieval and request a new retrieval
directed at the right color concept. Since memory retrievals take time, responses to
incongruent stimuli take longer than responses to neutral stimuli. In the congruent
Manuscript
condition, both control units spread activation toward the same concept in memory
increasing its activation and speeding up its retrieval (Figure 2). In addition, for
congruent stimuli, the first retrieval is sufficient for generating a correct response, even
when it is guided solely by the word dimension of the stimulus. Thus, facilitation is not
simply the reverse of interference, as noticed by (MacLeod & MacDonald, 2000). This
way of modeling within-trial effects is similar in principle with other models of the
Stroop task (Altmann & Davidson, 2001; Cohen, Dunbar, & McClelland, 1990; Herd,
Banich, & O'Reilly, 2006; Lovett, 2005; Roelofs, 2003). Notice that it does not require a
mechanism of suppression of the more automatic response in favor of the less automatic
but task relevant response. Such a suppression mechanism is only needed to account for
between-trial effects, as shown in the following sections.
(Figure 2 about here)
4.2. Bottom-up suppression (reactive inhibition)
Repetition avoidance has been extensively studied in cognitive control tasks such as the
task of generating sequences of random numbers (RNG). It seems that the process is
relatively automatic and does not rely on a limited capacity resource (Baddeley, Emslie,
Kolodny, & Duncan, 1998; Shallice, 2004). In addition, models of cognitive control in
sequential behavior often postulate a biphasic pattern of activation and suppression. In
short, this biphasic pattern consists of early activation followed by late suppression,
which should allow activation at novel locations or objects etc. (Klein, 2004; Pratt, Hillis,
Manuscript
& Gold, 2001; Tipper, Weaver, Jerreat, & Burak, 1994) According to this idea,
suppression follows activation in order to allow proper composition of sequences of
actions (Houghton & Tipper, 1996). Reactive inhibition is a related concept, which
claims that inhibition is greater to the extent that a distractor might be expected to
intrude. Reactive inhibition seems to be an aftereffect of processing that is not usually
intended (Logan, 1994). The adaptive function of a bottom-up suppression mechanism is
best explained in the following quotation:
“Many natural systems reflect a tendency for positive priming, such that an item
that has recently occurred is more readily accessed, and therefore if the system is
to avoid becoming locked in a positive feedback cycle of perseveration, there
needs to be some form of short-term and automatic inhibition or negative
priming.” (Baddeley et al. 1998, pp. 846)
ACT-R uses a form of inhibitory tagging (Fuentes, 1999; Jonides & Smith, 1997) to
implement inhibition of return effects in vision and to prevent perseverative retrieval in
memory tasks. An attended location in a visual display is tagged as attended and the
search for a new location to attend is biased toward locations that have not been tagged as
attended. Similarly, a retrieved memory element is tagged as recently-retrieved and a
new retrieval is biased toward memory elements that have not been tagged as recentlyretrieved. Tags are attached to memory elements for a while and eliminated after a
certain time has passed. This mechanism is called FINST (fingers of instantiation) and its
principles are borrowed from (Pylyshyn, 2000).
Manuscript
For our purposes, the memory-finst mechanism seems appropriate to model bottom-up
repetition suppression, since what is repeated is the semantic concept that underlies the
visual features of the stimuli. For example, in the Color-Word effect, there is no
repetition of any visual feature of the stimuli, but there is repetition of the concept that is
instantiated first as a color and then as a word. Representations of concepts in memory
are activated when the stimuli are perceived and one of these representations is needed
for naming the color of the current stimulus. Retrieving the correct concepts from
memory is the key toward generating correct responses.
A small adjustment to the standard memory-finst mechanism of ACT-R was necessary.
First, all-or-none tags attached to memory elements, as it is the case in the standard ACTR, were counterproductive. They would completely block a recently retrieved memory
element from being re-retrieved. However, the observed between-trial effects suggest that
re-retrieval is only delayed not blocked. Thus the finsts have been assigned a continuous
value instead of an all-or-none value. This value is subtracted from the activation of the
corresponding memory element and thus it slows down its retrieval, instead of blocking
it. Second, the empirical data suggest a gradual decay of this continuous value (see Figure
1), instead of a constant delay, as it is the case in the standard ACT-R. This adjustment
will be referred to as “the decaying finst mechanism”.
By adding a decaying finst mechanism to the model described in section 4.1 the pattern
of between trial effects shown in Figure 3 has been obtained. Thus, in the Word-Color
(W-C) trials, the concept corresponding to the word feature of the preceding stimulus has
Manuscript
been retrieved and finsted, that is, its activation has been discounted (The form “finsted”
has also been used by the author of the FINST concept; Pylyshyn, 2000). When the same
concept needs to be re-retrieved to name the color of the current stimulus, retrieval takes
longer than in control cases. When the repetition is not immediate but occurs over two or
three trials (W2-C and W3-C, respectively), the effect gets gradually smaller in
magnitude because the finst has decayed.
(Figure 3 about here)
In the Color-Word trials, the concept corresponding to the color feature of the preceding
stimulus has been retrieved and finsted. The same concept is associated with the word
feature of the current stimulus. Normally, this concept would have high activation due to
its high strength of association with the word control unit from the goal buffer and would
interfere with color naming on the current trial. However, because it has been finsted, this
concept is less likely to be retrieved in the current trial, that is, has less potential to
interfere with color naming on this trial. This explains the decrease in reaction time for
Color-Word trials.
In the Color-Color trials, the concept corresponding to the color feature of the preceding
stimulus has been retrieved and finsted. The same concept is associated with the color
feature of the current stimulus and it takes longer to be re-retrieved in order to be used in
naming the color. In the Word-Word trials, the concept corresponding to the word feature
of the preceding stimulus has been retrieved and finsted. The same concept is associated
Manuscript
with the word feature of the current stimulus. Normally, this concept would interfere with
color naming on the current trial, but due to its finsted activation it is less likely to be
retrieved, thus allowing a faster color naming than in control trials. However, in the
congruent condition, retrieval of this concept would normally produce facilitation, which
is now prevented from occurring because the concept has been finsted. This explains the
increase in reaction time in the Word-Word trials in the congruent condition as observed
in the empirical data (see table 3).
This model produces a reasonably good fit to the empirical data (Correlation = 0.924,
Mean deviation = 15.4 ms) by implementing a relatively simple mechanism – decaying
finst. This mechanism seems to implement well the main characteristics of a bottom-up
repetition suppression account, that is, it automatically applies to all memories, it is a
short-term aftereffect of activation, and it serves the function of preventing positive
feedback and perseveration. However, as shown in Figure 3, the model produces smaller
magnitudes for the Word-Color effect and larger magnitudes for the Color-Word and
Color-Color effects than observed in the empirical data. These deviations are caused by
an intrinsic characteristic of the decaying-finst mechanism, that is, it acts after retrieval.
In terms of (Logan, 1994), the suppression modeled by the decaying-finst mechanism is
an aftereffect of activation. In the terms of our ACT-R model, a memory element gets
finsted only if and right after it has been retrieved. Because the finst decays, the exact
moment of retrieval determines the magnitude of the aftereffect. Thus in Word-Color
trials, the concept corresponding to the word feature of the preceding stimulus was
retrieved before the concept corresponding to the color feature of the preceding stimulus,
Manuscript
because of its higher strength of association with the word control unit. By the time when
the same concept needs to be re-retrieved to name the color of the current stimulus the
finst has already decayed to some extent. This is why the Word-Color effects are smaller
in magnitude than expected. When the concept corresponding to the color feature of the
preceding stimulus repeats, as in the Color-Word and Color-Color effects, the effects are
larger because the retrieval and the subsequent finst have happened more recently than in
the case of repeating the concept corresponding to the word feature. Thus, these local
misfits are caused by the sequential order of processing for the word and color
dimensions of the stimulus. However, there is evidence that the two dimensions are
processed in parallel (MacLeod & Bors, 2002). If the two concepts were simultaneously
retrieved and then finsted, these misfits would have probably not occurred. Similar
problems with the
4.3. Top-down suppression (active inhibition)
Performance in the Stroop task usually correlates with performance in tasks that are
known to involve top-down suppression (Miyake et al., 2000), such as the antisaccade
task (Hallett, 1978) and stop-signal task (Logan, 1994). Stroop effects have been shown
to reverse, diminish or disappear in a number of circumstances (Harrison & Espelid,
2004; MacLeod, 1991; MacLeod & Sheehan, 2003; Metzler & Parkin, 2000). In
particular, negative priming (the Word-Color effect) occurs only under specific task
instructions (MacLeod, 1991) after extensive practice with a small and homogeneous set
of stimuli (Weger & Inhoff, 2006). These findings suggest that between-trial effects
Manuscript
might be caused by the same top-down suppression mechanism involved in stop signal
tasks (Aron et al., 2007; Verbruggen, Logan, Liefooghe, & Vandierendonck, in press).
A model implementing a top-down suppression mechanism has been developed as an
alternative to the decaying finst model. This model postulates a control structure that is
external to the memory retrieval process and only influences it in particular
circumstances. This structure, called “the suppression buffer”, is analogous to the goal
buffer in that it has a biasing influence on memory retrieval, except that this influence is
opposite in sign. The activation spread from the goal buffer has an excitatory influence
on the targeted memory elements, whereas the activation spread from the suppression
buffer has an inhibitory influence on the targeted memory elements. The goal buffer and
the suppression buffer are independent from each other, that is, they can operate
simultaneously on the same memory element. In other words, suppression is not an
aftereffect of activation. Another important difference with the decaying finst model is
that suppression is not a property of a memory element (a quantity that is attached to it)
but it is an external influence that lasts as long as the source of this influence lasts.
This model assumes that top-down suppression is employed in order to prevent
information from preceding trials from interfering with the processing required by the
current trial. After a trial has been completed its associated stimulus is represented in the
suppression buffer. The two features of this stimulus (word and color) spread negative
activation (i.e., suppression) toward their associated concepts represented in memory.
When the suppressed concepts need to be retrieved to assist with the processing of the
Manuscript
current stimulus, the known between-trial effects are manifested. The suppression buffer
can accommodate up to three preceding stimuli and their source of suppression decays
gradually. Buffers that accommodate multiple representations with different decaying
sources of spreading activation are not supported by the standard ACT-R architecture.
Negative spreading activation is usually not used in ACT-R models (although, see Van
Maanen and Van Rijn, 2007, for an exception). These features were created ad-hoc for
this model in order to illustrate the theoretical value of a top-down suppression account.
Running the top-down suppression model produced the pattern of results presented in
Figure 4. This model also produces a good fit to the data (Correlation = 0.949, Mean
deviation = 11.4 ms), even a slightly better fit than the decaying finst model. The four
between-trial effects have now comparable magnitudes. However, the Color-Color effect
in the empirical data is not as large as the model estimates it. Perhaps the magnitude of
the Color-Color effect is also modulated by other factors that we have not considered in
our model.
(Figure 4 about here)
5. General discussion and conclusion
Criticism has recently been expressed with reference to psychological theories that
postulate suppression (cognitive inhibition) as an explanatory mechanism for observed
behavioral effects (MacLeod, Dodd, Sheard, Wilson, & Bibi, 2003). Since we agreed that
Manuscript
many of these points of criticism were justified, we seriously considered them while
conducting the research reported here. One of these criticisms states that the term
cognitive inhibition is misleading because it creates confusion with the phenomenon of
neural inhibition. In response to this criticism, we have adopted the term “suppression”
instead of “cognitive inhibition”. Another criticism points at a tendency to postulate
suppression for any findings showing decreases in performance before alternative
suppression-free accounts have been considered. We have addressed known behavioral
effects showing both increases and decreases in performance and tried to explain them
with an integrated account. Before postulating a suppression account, we have analyzed
the existing suppression-free accounts, and showed that they fail to explain all effects in
an integrated way.
Although the between-trial effects presented here have been known for a long time (see
MacLeod, 1991, for a review), we have replicated all of them in the same study and have
been able to estimate their time course and relative magnitudes while controlling for
within-trial effects. An integrated account for these effects has not been put forward so
far according to our knowledge. We have shown that a repetition suppression account can
explain all these effects and have demonstrated two neurally plausible ways to implement
this account in a computational cognitive model. The neural substrate of bottom-up
(reactive) suppression could be the circuitry between locus coeruleus (LC) and cortex.
Neural activity is increased in LC and its projections to cortex during processing of taskrelevant stimuli. After processing of a particular stimulus, LC inhibits itself via local
connections, creating a refractory period in which the cortex is unable to process the
Manuscript
same stimulus or similar stimuli (Nieuwenhuis, Gilzenrat, Holmes, & Cohen, 2005). The
neural substrate of top-down suppression could be the fronto-subthalamic circuitry, also
called the indirect pathway (Aron, 2007; Chambers et al., 2006). The top-down
suppression signal is originated in the right inferior frontal cortex and activates the
subthalamic nucleus (STN). STN excites globus pallidum and this inhibits thalamus. As a
result, thalamus stops activating specific cortical areas.
The computational mechanism that we have used to model bottom-up suppression
(decaying-finst) is in line with the way the ACT-R theory models suppression in memory
and vision phenomena. We have only added a decaying characteristic to the classical finst
mechanism of ACT-R in order to account for the observed gradual decay of between-trial
effects. The mechanism we have used to model top-down suppression (decaying negative
spreading activation) is relatively novel and it needs more research in order to be fully
validated. Although evidence in favor of this account from fMRI and TMS studies tends
to accumulate (Aron, 2007; Chambers et al., 2006), there are many aspects that need
further investigation.
In an upcoming study, we will test the two accounts more directly. We will try to separate
the activation process from the retrieval process by presenting extra stimuli next to the
target stimulus. The task instructions will emphasize that the extra stimuli will have to be
ignored. In such conditions, the two models make different predictions. The top-down
model predicts that traces of suppression (negative priming) of the extra stimuli will be
detected, since suppression will explicitly be required by the task instructions. The
Manuscript
bottom-up model predicts no such traces of suppression for the extra-stimuli, since their
corresponding representations will merely be activated but not retrieved, thus not-finsted.
Instead, the bottom-up model predicts traces of activation (positive priming) produced by
the extra stimuli.
In conclusion, we have attempted to demonstrate that a repetition-suppression mechanism
is a viable theoretical account for explaining a range of between-trial effects in an
integrated way. More research is needed to adequately characterize this account and in
particular to answer the question whether repetition suppression is an intrinsic property of
memory activation or it is the effect of a controlling influence that is external to the
activation process and it only acts in specific circumstances.
Manuscript
References
Altmann, E. M., & Davidson, D. J. (2001). An integrative approach to Stroop:
Combining a language model and a unified cognitive theory Paper presented at
the Twenty-Third Annual Conference of the Cognitive Science Society, Hillsdale,
NJ
Anderson, J. R. (2007). How Can the Human Mind Occur in the Physical Universe? New
York: Oxford University Press.
Aron, A. R. (2007). The Neural Basis of Inhibition in Cognitive Control Neuroscientist,
13(3), 214-228.
Aron, A. R., Durston, S., Eagle, D. M., Logan, G. D., Stinear, C. M., & Stuphorn, V.
(2007). Converging Evidence for a Fronto-Basal-Ganglia Network for Inhibitory
Control of Action and Cognition. The Journal of Neuroscience 27(44), 1186011864.
Baddeley, A. D., Emslie, H., Kolodny, J., & Duncan, J. (1998). Random Generation and
the Executive Control of Working Memory. The Quarterly Journal of
Experimental Psychology: Section A, 51(4), 819-852.
Cattell, J. M. (1886). The time it takes to see and name objects. Mind, 11, 63-65.
Chambers, C. D., Bellgrove, M. A., Stokes, M. G., Henderson, T. R., Garavan, H.,
Robertson, I. H., et al. (2006). Executive "Brake Failure" following Deactivation
of Human Frontal Lobe. Journal of Cognitive Neuroscience, 18(3), 444 - 455.
Christie, J., & Klein, R. M. (2001). Negative priming for spatial location? Can J Exp
Psychol, 55(1), 24-38.
Manuscript
Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of automatic
processes: A parallel distributed processing account of the Stroop effect.
Psychological Review, 97, 332-361.
Effler, M. (1977). The influence of serial factors on the Stroop test. Psychologische
Beitrage, 19, 189-200.
Egner, T., & Hirsch, J. (2005). Cognitive control mechanisms resolve conflict through
cortical amplification of task-relevant information. Nature Neuroscience, 8(17841790).
Erickson, M. A., & Reder, L. M. (1998). The Influence of Repeated Presentations and
Intervening Trials on Negative Priming Paper presented at the The Twentieth
Annual Conference of the Cognitive Science Society Mahwah, NJ
Fraisse, P. (1969). Why is naming longer than reading? . Acta psychol., 30, 96-103.
Fuentes, L. J. (1999). Inhibitory Tagging of Stimulus Properties in Inhibition of Return:
Effects on Semantic Priming and Flanker Interference The Quarterly Journal of
Experimental Psychology: Section A, 52(1), 149 - 164.
Fuentes, L. J., Boucart, M., Vivas, A. B., Alvarez, R., & Zimmerman, M. A. (2000).
Inhibitory tagging in inhibition of return is affected in schizophrenia: evidence
from the stroop task. Neuropsychology, 14(1), 134-140.
Fuentes, L. J., Vivas, A. B., & Humphreys, G. W. (1999). Inhibitory mechanisms of
attentional networks: Spatial and semantic inhibitory processing Journal of
Experimental Psychology: Human Perception and Performance, 25, 1114-1126.
Hallett, P. E. (1978). Primary and secondary saccades to goals defined by instructions.
Vision Research, 18, 1279–1296.
Manuscript
Harrison, Y., & Espelid, E. (2004). Loss of negative priming following sleep deprivation.
The Quarterly Journal of Experimental Psychology: Section A, 57(3), 437-446.
Herd, S. A., Banich, M. T., & O'Reilly, R. C. (2006). Neural Mechanisms of Cognitive
Control: An Integrative Model of Stroop Task Performance and fMRI Data J.
Cognitive Neuroscience MIT Press, 18(1 ), 22-32
Houghton, G., & Tipper, S. P. (1996). Inhibitory Mechanisms of Neural and Cognitive
Control: Applications to Selective Attention and Sequential Action Brain and
Cognition, 30, 20-43.
Houghton, G., Tipper, S. P., Weaver, B., & Shore, D. I. (1996). Inhibition and
interference in selective attention: Some tests of a neural network model. Visual
Cognition, 3(119–164).
Jonides, J., & Smith, E. E. (1997). The architecture of working memory. In M. D. Rugg
(Ed.), Cognitive Neuroscience (pp. 243-276). Sussex, England: Psychology Press.
Klein, R. M. (2004). Orienting and Inhibition of Return. In M. S. Gazzaniga (Ed.), The
Cognitive Neurosciences III. Cambridge, Massachusetts: MIT Press.
Law, M. B., Pratt, J., & Abrams, R. A. (1995). Color-based inhibition of return. Percept
Psychophys., 57(3), 402-408.
Logan, G. D. (1990). Repetition priming and automaticity: Common underlying
mechanisms? Cognitive Psychology, 22, 1-35.
Logan, G. D. (1994). On the ability to inhibit thought and action: A users' guide to the
stop signal paradigm. In D. Dagenbach & T. H. Carr (Eds.), Inhibitory processes
in attention, memory, and language. San Diego: Academic Press.
Manuscript
Lovett, M. C. (2005). A Strategy-Based Interpretation of Stroop. Cognitive Science(29),
493-524.
MacLeod, C. M. (1991). Half a Century of Research on the Stroop Effect: An Integrative
Review Psychological Bulletin, 109(2), 163-203.
MacLeod, C. M. (2007a). Cognitive inhibition: Elusive or illusion? In I. H. L. Roediger,
Y. Dudai & S. M. Fitzpatrick (Eds.), Science of memory: Concepts (pp. 301-305).
New York: Oxford University Press.
MacLeod, C. M. (2007b). The concept of inhibition in cognition. In D. S. Gorfein & C.
M. MacLeod (Eds.), Inhibition in cognition (pp. 3-23 ). Washington, DC:
American Psychological Association.
MacLeod, C. M., & Bors, D. A. (2002). Presenting two color words on a single Stroop
trial: Evidence for joint influence, not capture Memory and Cognition, 30(5), 789797.
MacLeod, C. M., Dodd, M. D., Sheard, E. D., Wilson, D. E., & Bibi, U. (2003). In
Opposition to Inhibition. In B. H. Ross (Ed.), The Psychology of Learning and
Motivation (Vol. 43, pp. 163-214): Elsevier Science.
MacLeod, C. M., & MacDonald, P. A. (2000). Inter-dimensional interference in the
Stroop effect: Uncovering the cognitive and neural anatomy of attention. Trends
in Cognitive Sciences, 4, 383-391.
MacLeod, C. M., & Sheehan, P. W. (2003). Hypnotic control of attention in the Stroop
task: A historical footnote. Consciousness and Cognition 12(3), 347-353.
Metzler, C., & Parkin, A. J. (2000). Reversed negative priming following frontal lobe
lesions. Neuropsychologia, 38, 363-379.
Manuscript
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T.
D. (2000). The unity and diversity of executive functions and their contributions
to complex "frontal lobe" tasks: A latent variable analysis. Cognitive Psychology,
41, 49-100.
Neill, W. T. (1997). Episodic Retrieval in Negative Priming and Repetition Priming.
Journal of Experimental Psychology; Learning, Memory, and Cognition 23(6),
1291-1305.
Neill, W. T., & Westberry, R. L. (1987). Selective attention and the suppression of
cognitive noise. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 13, 327-334.
Nieuwenhuis, S., Gilzenrat, M. S., Holmes, B. D., & Cohen, J. D. (2005). The Role of the
Locus Coeruleus in Mediating the Attentional Blink: A Neurocomputational
Theory. Journal of Experimental Psychology: General, 134(3), 291–307.
Park, J., & Kanwisher, N. (1994). Negative priming for spatial locations: Identity
mismatching, not distractor inhibition. Journal of Experimental Psychology:
Human Perception and Performance, 20, 613-623.
Posner, M. I., & Cohen, Y. (1984). Components of visual attention. In H. Bouma & D. G.
Bouwhuis (Eds.), Attention & Performance (Vol. 10, pp. pp. 531 - 556). Hillsdale,
NJ: Erlbaum.
Pratt, J., Hillis, J., & Gold, J. M. (2001). The effect of the physical characteristics of cues
and targets on facilitation and inhibition. Psychonomic Bulletin & Review, 8(3),
489-495.
Manuscript
Pratt, J., Spalek, T. M., & Bradshaw, F. (1999). The time to detect targets at inhibited and
noninhibited locations: Preliminary evidence for attentional momentum. Journal
of Experimental Psychology: Human Perception and Performance, 25(730–746).
Pylyshyn, Z. W. (2000). Situating vision in the world Trends in Cognitive Sciences, 4(5),
197-207.
Roelofs, A. P. A. (2003). Goal-referenced selection of verbal action: Modeling attentional
control in the Stroop task Psychological review, 110, 88-124.
Shallice, T. (2004). The fractionation of supervisory control. In M. S. Gazzaniga (Ed.),
The Cognitive Neurosciences III. Cambridge, Massachusetts MIT Press.
Thomas, J. K. (1977). Stroop interference with word or hue pre-exposure. University of
Michigan.
Tipper, S. P. (2001). Does negative priming reflect inhibitory mechanisms? A review and
integration of conflicting views. The Quarterly Journal of Experimental
Psychology: Section A, 54A, 321–343.
Tipper, S. P., Weaver, B., Cameron, S., Brehaut, J. C., & Bastedo, J. (1991). Inhibitory
mechanisms of attention in identification and localization tasks: Time course and
disruption. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 17(4), 681-692.
Tipper, S. P., Weaver, B., Jerreat, L. M., & Burak, A. L. (1994). Object-based and
environment-based inhibition of return of visual attention. J Exp Psychol Hum
Percept Perform, 20(3), 478-499.
Van Maanen, L., & Van Rijn, H. (2007). An Accumulator Model of Semantic
Interference. Cognitive Systems Research, 8(3), 174-181.
Manuscript
Verbruggen, F., Logan, G. D., Liefooghe, B., & Vandierendonck, A. (in press). Shortterm aftereffects of response inhibition: Repetition priming or between-trial
control adjustments? Journal of Experimental Psychology: Human Perception
and Performance.
Vivas, A. B., & Fuentes, L. J. (2001). Stroop interference is affected in inhibition of
return. Psychon Bull Rev, 8(2), 315-323.
Weger, U. W., & Inhoff, A. W. (2006). Semantic inhibition of return is the exception
rather than the rule. Perception and Psychophysics, 68, 244-253.
Manuscript
Table 1. Typical within-trial effects: the reaction time (RT) in the incongruent and
congruent conditions is higher and lower, respectively, than in the neutral condition
(interference and facilitation, respectively). The inverse pattern occurs for accuracy data.
Incongruent
Congruent
Neutral
Accuracy (%)
0.92
0.99
0.98
Mean RT (ms)
1183.6
953.1
1055
Manuscript
Table 2. A synoptic with examples of between-trial effects
Preceding trial
Current trial
Reaction time
Word-Color
RED
GREEN
Increase
Color-Word
YELLOW
RED
Decrease
Word-Word
BLUE
BLUE
Decrease
Color-Color
RED
BLUE
Increase
Manuscript
Table 3. The absolute magnitudes (ms) of the between-trial effects per condition
Incongruent
Congruent
Neutral
Word-Color
102
93
40
Color-Word
-74
8
---
Word-Word
-68
93
-51
Color-Color
79
8
23
Manuscript
Table 4. Results of the initial LME model
Value
Std.Error
DF
t-value
p-value
(Intercept)
1197.1
27.4
16140
43.7
0.000
Congruent Condition
-235.9
10.5
16140
-22.4
0.000
Neutral Condition
-156.3
10.9
16140
-14.3
0.000
Word-Color
50.7
12.9
16140
3.9
0.000
Color-Word
-37.1
12.5
16140
-3.0
0.003
Color-Color
31.7
10.1
16140
3.1
0.002
Word-Word
-39.7
19.3
16140
-2.1
0.039
Word-Color-2back
29.1
11.3
16140
2.6
0.010
Color-Word-2back
-26.3
12.0
16140
-2.2
0.028
Color-Color-2back
16.4
9.5
16140
1.7
0.085
Word-Word-2back
-16.7
13.7
16140
-1.2
0.222
Word-Color-3back
8.0
11.1
16140
0.7
0.473
Color-Word-3back
-7.4
12.0
16140
-0.6
0.539
Color-Color-3back
7.4
9.7
16140
0.8
0.445
Word-Word-3back
2.4
13.7
16140
0.2
0.863
Congr*Word-Word
99.6
33.3
16140
3.0
0.003
Neutral*Word-Word
25.9
55.8
16140
0.5
0.642
Manuscript
Table 5. Results of the best fitting LME model
Value
Std.Error
DF
t-value
p-value
(Intercept)
1197.7
27.0
16145
44.3
0.000
Cond-congruent
-236.2
10.5
16145
-22.5
0.000
Cond-neutral
-153.5
10.4
16145
-14.7
0.000
Word-Color
50.9
12.9
16145
4.0
0.000
Color-Word
-37.4
12.5
16145
-3.0
0.003
Color-Color
31.6
10.1
16145
3.1
0.002
Word-Word
-39.6
19.3
16145
-2.1
0.040
Word-Color-2back
24.9
10.7
16145
2.3
0.020
Color-Word-2back
-30.1
11.5
16145
-2.6
0.009
Color-Color-2back
17.2
9.5
16145
1.8
0.070
Congr*Word-Word
99.0
33.3
16145
3.0
0.003
Neutral*Word-Word
26.4
55.8
16145
0.5
0.636
Manuscript
Figure 1. LME Coefficients for the between-trial effects. Stars above some of the bars
indicate significant effects (alpha=0.05).
Manuscript
Figure 2. Within-trial effects as shown in the empirical data and in the two models
presented in sections 4.2 and 4.3.
Manuscript
Figure 3. The pattern of between-trial effects as estimated from empirical data and
simulated by the “decaying finst” model. On the horizontal axis the four between-trial
effects and their corresponding 2- and 3-back variants are deployed. The vertical axis
shows the magnitudes of these effects: positive values indicate increase and negative
values decrease in reaction time as compared to control trials. The error bars indicate
standard errors of the means.
Manuscript
Figure 4. The pattern of between-trial effects as estimated from empirical data and
simulated by the “top-down suppression” model. On the horizontal axis the four betweentrial effects and their corresponding 2- and 3-back variants are deployed. The vertical axis
shows the magnitudes of these effects: positive values indicate increase and negative
values decrease in reaction time as compared to control trials. The error bars indicate
standard errors of the means.